System.Runtime.Extensions.Tests: Assertion failed #61359
Assignees
Labels
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
blocking-clean-ci
Blocking PR or rolling runs of 'runtime' or 'runtime-extra-platforms'
Comments
|
Tagging subscribers to this area: @dotnet/area-system-numerics Issue DetailsFailed in: #61323 Click to expand log
|
dotnet-issue-labeler
bot
added
the
untriaged
|
cc: @tannergooding |
|
(Libraries Test Run checked coreclr Linux x64 Debug) Guess - it's one of the Half values used as theory data (see BitConverterBase.cs) as it's constructing the test name. |
|
Failed again in #64851 on linux x64 - console log. Intermittent nature of failure suggests that perhaps some unsafe API operation is not reliable? 🤔 I tried |
|
We should break this assert into several, and log the values (including the half value.. carefully.) |
Ah, that's the key; otherwise we can potentially run into infinite recursion here (if debug message has Half value). |
I doubt that it would help. It has very high chance to make the repro go away. This is likely either going to be uninitialized variable somewhere or a codegen bug. I will take a look at the crash dump. |
|
This is a codegen bug in OptimizedTier1 version of The bug manifest itself as test failure only when we switch from QuickJitted version of More details based on the dump from #64851: |
jkotas
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT Issue DetailsFailed in: #61323 Click to expand log
|
jkotas
added
blocking-clean-ci
|
@dotnet/jit-contrib @JulieLeeMSFT This intermittent bug is affecting CI stability. Could you please give it a high priority? |
|
This sounds suspiciously much like #58373, I will take a look. |
|
Thanks for the repro! Standalone repro: using System;
using System.Runtime.CompilerServices;
using System.Threading;
public class Test
{
public static void Main()
{
for (int i = 0; i < 100; i++)
{
Console.WriteLine(HalfToInt16Bits((Half)(-3.14f)));
Thread.Sleep(16);
}
}
[MethodImpl(MethodImplOptions.NoInlining)]
public static unsafe short HalfToInt16Bits(Half value)
{
return *((short*)&value);
}
}
|
jakobbotsch
added a commit
to jakobbotsch/runtime
that referenced
this issue
Feb 6, 2022
This ensures we normalize returned indirs of locals, even when we fold it into an access of a promoted local's field. This may change a signed indir into access of an unsigned field, which may need a sign-extending cast to be inserted. It is not ideal that fgMorphRetInd can 'lose' this information, but given that it is a specialized optimization it seems the simplest solution is to just rely on the follow-up normalization. Fix dotnet#61359
ghost
added
the
in-pr
jakobbotsch
removed
the
untriaged
jakobbotsch
added a commit
that referenced
this issue
Feb 7, 2022
This ensures we normalize returned indirs of locals, even when we fold it into an access of a promoted local's field. This may change a signed indir into access of an unsigned field, which may need a sign-extending cast to be inserted. It is not ideal that fgMorphRetInd can 'lose' this information, but given that it is a specialized optimization it seems the simplest solution is to just rely on the follow-up normalization. Fix #61359
ghost
removed
the
in-pr
ghost
locked as resolved and limited conversation to collaborators
github-actions bot
pushed a commit
that referenced
this issue
Mar 1, 2023
This ensures we normalize returned indirs of locals, even when we fold it into an access of a promoted local's field. This may change a signed indir into access of an unsigned field, which may need a sign-extending cast to be inserted. It is not ideal that fgMorphRetInd can 'lose' this information, but given that it is a specialized optimization it seems the simplest solution is to just rely on the follow-up normalization. Fix #61359
carlossanlop
pushed a commit
that referenced
this issue
Mar 10, 2023
* Normalize returns after optimizing returned indirs This ensures we normalize returned indirs of locals, even when we fold it into an access of a promoted local's field. This may change a signed indir into access of an unsigned field, which may need a sign-extending cast to be inserted. It is not ideal that fgMorphRetInd can 'lose' this information, but given that it is a specialized optimization it seems the simplest solution is to just rely on the follow-up normalization. Fix #61359 * Conservative fix * Remove spurious diff --------- Co-authored-by: Jakob Botsch Nielsen <jakob.botsch.nielsen@gmail.com>
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Failed in: #61323
Click to expand log
The text was updated successfully, but these errors were encountered: